Cognition in funding talks at $25B valuation — Devin + Windsurf becoming the standalone harness empire

What’s reported and what’s confirmed

The reporting from Bloomberg, Yahoo Finance, SiliconAngle, and TechFundingNews:

• Talks ongoing for a new funding round at a $25 billion valuation
• Raise size: hundreds of millions of dollars (specific number not disclosed)
• Prior round: $400M led by Founders Fund in September 2025, at $10.2B post-money
• Status: talks ongoing; terms could shift before close

What’s already closed and confirmed:

• July 2025: Cognition signed definitive agreement to acquire Windsurf
• December 2025: Windsurf acquisition closed for approximately $250M
• Context: just days before Cognition’s announcement, Google reverse-acquihired Windsurf’s CEO Varun Mohan, co-founder Douglas Chen, and research leaders for $2.4B — leaving the 250-person team behind for Cognition to inherit

The financial trajectory

Devin and Windsurf as separate financial entities (pre-merger):

Devin’s growth rate: $1M → $73M in nine months is a 73× annualized multiple. For context, that’s faster than Claude Code’s 2025 trajectory was at the equivalent stage.

Post-merger figures from the most recent reporting:

• Combined enterprise ARR rose 30%+ in the seven weeks immediately following the Windsurf close
• The acquisition was reported as “immediately accretive” by Cognition

The SWE-1.5 angle

A detail most coverage glosses over but is structurally important: Cognition isn’t running Devin + Windsurf on Anthropic / OpenAI / Google models. The company built its own proprietary SWE-1.5 model, reportedly running at 950 tokens/second — roughly 13× faster than Claude Sonnet 4.5 on equivalent coding tasks.

This matters because most of the AI harnesses category is downstream of three model labs (Anthropic, OpenAI, Google). Claude Code depends on Claude; Antigravity depends on Gemini; OpenAI Codex depends on GPT. Cognition is the first major harness vendor with its own coding-specialized model, which means:

1. No vendor risk if Anthropic raises prices
2. Coding-task latency optimized end-to-end (the 950-token/sec figure)
3. Margin captured in-house rather than passed to API providers
4. The harness is the model business — vertically integrated

This is closer to the OpenAI / Anthropic model than to the Cursor / Aider model. Cognition isn’t an API consumer; it’s a model lab plus harness plus IDE in one company.

What it means for the harness category

The AI harnesses category was, until this story, structurally divided into:

• Big-tech harnesses (Claude Code, Codex, Antigravity, GitHub Copilot)
• Model-agnostic open-source (OpenCode, Aider, Pi, Goose)
• IDE specialists (Cursor, Windsurf)
• Autonomous specialists (Devin)

Cognition’s combined Devin + Windsurf + SWE-1.5 stack cuts across three of those four buckets. It’s a big-tech-scale harness business that wasn’t built by a model lab. The $25B valuation reflects investor consensus that:

1. The harness layer doesn’t need to be owned by a model lab to capture meaningful revenue
2. Vertical integration (model + harness + IDE) is a defensible moat
3. The market is large enough to support a multi-billion-dollar standalone harness company

If the round closes, Cognition becomes the fourth most valuable AI dev-tools company (behind GitHub-owning Microsoft, OpenAI’s IPO-bound stack, and Anthropic’s Claude Code business). For an 18-month-old company that’s a substantial sorting.

What it means for Devin and Windsurf users

Practically, very little in the immediate term. Both products continue to operate separately with their own UIs. The shared SWE-1.5 backend is already in production.

What changes structurally:

• For Windsurf users: the product survives. Pre-acquisition concerns about Codeium’s future are moot — Cognition has both the capital and the strategic intent to keep developing it
• For Devin users: the financial backing for continued capability investment is now substantially stronger
• For both: expect tighter integration over the next 12-18 months; Devin’s autonomous capabilities will likely become accessible from inside Windsurf, and vice versa

The honest caveats

Two caveats worth surfacing:

The round isn’t closed. “Funding talks at $25 billion” is reporting-stage information. Terms can shift. AI valuation rounds at this scale occasionally collapse or restructure before close. Bloomberg’s reporting is solid but treat the $25B as the expected close valuation rather than the confirmed one.

Devin’s autonomous-SWE pitch remains controversial. Cognition’s marketing has at times overstated Devin’s autonomous capabilities (the original 2024 SWE-Bench demo was scrutinized for cherry-picking). The $25B valuation reflects strong revenue growth, but the “will replace engineers” framing should be read with healthy skepticism.

What it changes for Pick Right readers tomorrow

If you’re a Devin subscriber, nothing changes about your service. If you’re a Windsurf user, same. What changes is the structural context: Cognition is now sized to be a permanent fixture in the harness category, not a startup that could be acquired or shut down.

For broader context, see the Devin review, the Windsurf review, and the best AI harnesses roundup. For the broader 2026 AI funding picture, see the Anthropic $30B raise coverage and the OpenAI confidential S-1 filing.

Questions or corrections? Email Pick Right. Want the full list? See all news.